Page 75 - 《应该声学》2022年第2期
P. 75

第 41 卷 第 2 期                                                                       Vol. 41, No. 2
             2022 年 3 月                          Journal of Applied Acoustics                    March, 2022

             ⋄ 研究报告 ⋄



                一种基于特征点提取的扬声器异常声检测方法                                                                    ∗





                                           宋华建     1   穆瑞林      1,2†  周子奇     1


                                            (1  天津科技大学机械工程学院        天津   300222)
                                (2  天津市轻工与食品工程机械装备集成设计与在线监控重点实验室                天津   300222)

                摘要：针对短时傅里叶变换在扬声器异常声检测中有效信息提取的随机性问题，提出了特征点法在扬声器异
                常声检测中的应用。此方法基于扬声器经扫频信号激励所得响应信号的短时傅里叶变换时频图，用改进的尺
                度不变特征转换算法对合格扬声器与异常声扬声器做特征提取，并将多组特征点经分割剔除后叠加组成特征
                矩阵模板。以合格扬声器样本提取特征曲线阈值构建检测模型判断扬声器是否存在异常声故障，以不同故障
                类型扬声器的专有特征点进行故障分类。实验结果表明，此方法可有效提取扬声器异常声特征，故障样本检出
                率可达 97.63%，故障分类精度可达 95%。
                关键词：特征点；异常声；时频图；尺度不变特征转换
                中图法分类号: TN643           文献标识码: A          文章编号: 1000-310X(2022)02-0243-07
                DOI: 10.11684/j.issn.1000-310X.2022.02.009





                  Method for detecting rub & buzz of loudspeaker based on feature points


                                        SONG Huajian  1  MU Ruilin 1,2  ZHOU Ziqi 1

                     (1 College of Mechanical Engineering, Tianjin University of Science & Technology, Tianjin 300222, China)
             (2 Tianjin Key Laboratory of Integrated Design and On-line Monitoring for Light Industry & Food Machinery and Equipment,
                                    Tianjin University of Science & Technology, Tianjin 300222, China)

                 Abstract: Aiming at the randomness of effective information extraction by short-time Fourier transform
                 (STFT) in loudspeakers Rub & Buzz detection. The application of feature points method in loudspeaker Rub
                 & Buzz detection was proposed. This method was based on the time-frequency spectrum which was changed
                 from the sound signal by STFT. And the sound signal was emitted by the loudspeaker which was excited by a
                 sweep signal. Improved scale-invariant feature transform (SIFT) algorithm was used to extract feature points
                 from the time-frequency spectrum of the qualified loudspeakers and Rub & Buzz of loudspeakers. The feature
                 matrix can be obtained with removing the invalid points. The detection model can be constructed with the
                 qualified loudspeakers to test whether the loudspeakers were Rub & Buzz of loudspeaker. And the characteristic
                 points of loudspeakers with different fault types were used for fault classification. The experimental results
                 show that the proposed method can effectively extract the feature of loudspeaker Rub & Buzz. The detection
                 rate of fault samples can reach 97.63 percent. The accuracy of fault classification can reach 95 percent.
                 Keywords: Feature points; Rub & Buzz; Time-frequency spectrum; Scale-invariant feature transform


             2021-03-12 收稿; 2021-05-23 定稿
             天津市建委科技项目 (2017-10)
             ∗
             作者简介: 宋华建 (1995– ), 男, 天津人, 硕士研究生, 研究方向: 机械测试理论与技术。
              通信作者 E-mail: mrl3667@tust.edu.cn
             †
   70   71   72   73   74   75   76   77   78   79   80